Tutorial 2: Caratheodory’s Extension 1

2. Caratheodory’s Extension
In the following, Ω is a set. Whenever a union of sets is denoted  as
opposed to ∪, it indicates that the sets involved are pairwise disjoint.

Definition 6 A semi-ring on Ω is a subset S of the power set P(Ω)

with the following properties:
(i) ∅∈S
(ii) A, B ∈ S ⇒ A ∩ B ∈ S
n

(iii) A, B ∈ S ⇒ ∃n ≥ 0, ∃Ai ∈ S : A \ B = Ai
i=1

The last property (iii) says that whenever A, B ∈ S, there is n ≥ 0

and A1 , . . . , An in S which are pairwise disjoint, such that A \ B =
A1  . . .  An . If n = 0, it is understood that the corresponding union
is equal to ∅, (in which case A ⊆ B).

www.probability.net
Tutorial 2: Caratheodory’s Extension 2

Definition 7 A ring on Ω is a subset R of the power set P(Ω) with

the following properties:
(i) ∅∈R
(ii) A, B ∈ R ⇒ A ∪ B ∈ R
(iii) A, B ∈ R ⇒ A \ B ∈ R

Exercise 1. Show that A ∩ B = A \ (A \ B) and therefore that a

ring is closed under pairwise intersection.
Exercise 2.Show that a ring on Ω is also a semi-ring on Ω.
Exercise 3.Suppose that a set Ω can be decomposed as Ω = A1 
A2  A3 where A1 , A2 and A3 are distinct from ∅ and Ω. Define
 
S1 = {∅, A1 , A2 , A3 , Ω} and S2 = {∅, A1 , A2  A3 , Ω}. Show that S1
and S2 are semi-rings on Ω, but that S1 ∩ S2 fails to be a semi-ring
on Ω.
Exercise 4. Let (Ri )i∈I be an arbitrary family of rings on Ω, with

I = ∅. Show that R = ∩i∈I Ri is also a ring on Ω.

www.probability.net
Tutorial 2: Caratheodory’s Extension 3

Exercise 5. Let A be a subset of the power set P(Ω). Define:


R(A) = {R ring on Ω : A ⊆ R}
Show that P(Ω) is a ring on Ω, and that R(A) is not empty. Define:
 
R(A) = R
R∈R(A)

Show that R(A) is a ring on Ω such that A ⊆ R(A), and that it is

the smallest ring on Ω with such property, (i.e. if R is a ring on Ω
and A ⊆ R then R(A) ⊆ R).
Definition 8 Let A ⊆ P(Ω). We call ring generated by A, the
ring on Ω, denoted R(A), equal to the intersection of all rings on Ω,
which contain A.

Exercise 6.Let S be a semi-ring on Ω. Define the set R of all finite

unions of pairwise disjoint elements of S, i.e.

R = {A : A = ni=1 Ai for some n ≥ 0, Ai ∈ S}

www.probability.net
Tutorial 2: Caratheodory’s Extension 4

(where if n = 0, the corresponding union is empty, i.e. ∅ ∈ R). Let

A = ni=1 Ai and B = pj=1 Bj ∈ R:
1. Show that A ∩ B = i,j (Ai ∩ Bj ) and that R is closed under
pairwise intersection.
2. Show that if p ≥ 1 then A \ B = ∩pj=1 (ni=1 (Ai \ Bj )).
3. Show that R is closed under pairwise difference.
4. Show that A ∪ B = (A \ B)  B and conclude that R is a ring
on Ω.
5. Show that R(S) = R.

Exercise 7. Everything being as before, define:


R = {A : A = ∪ni=1 Ai for some n ≥ 0, Ai ∈ S}

www.probability.net
Tutorial 2: Caratheodory’s Extension 5

(We do not require the sets involved in the union to be pairwise dis-
joint). Using the fact that R is closed under finite union, show that
R ⊆ R, and conclude that R = R = R(S).
Definition 9 Let A ⊆ P(Ω) with ∅ ∈ A. We call measure on A,
any map μ : A → [0, +∞] with the following properties:
(i) μ(∅) = 0
+∞
 +∞

(ii) A ∈ A, An ∈ A and A = An ⇒ μ(A) = μ(An )
n=1 n=1

The  indicates that we assume the An ’s to be pairwise disjoint in

the l.h.s. of (ii). It is customary to say in view of condition (ii) that
a measure is countably additive.
Exercise 8.If A is a σ-algebra on Ω explain why property (ii) can
be replaced by:
+∞
 +∞

(ii) An ∈ A and A = An ⇒ μ(A) = μ(An )
n=1 n=1

www.probability.net
Tutorial 2: Caratheodory’s Extension 6

Exercise 9. Let A ⊆ P(Ω) with ∅ ∈ A and μ : A → [0, +∞] be a

measure on A.
1. Show that if A1 , . . . , An ∈ A are pairwise disjoint and the union
A = ni=1 Ai lies in A, then μ(A) = μ(A1 ) + . . . + μ(An ).
2. Show that if A, B ∈ A, A ⊆ B and B \A ∈ A then μ(A) ≤ μ(B).

Exercise 10. Let S be a semi-ring on Ω, and μ : S → [0, +∞] be a

measure on S. Suppose that there exists an extension of μ on R(S),
i.e. a measure μ̄ : R(S) → [0, +∞] such that μ̄|S = μ.
1. Let A be an element of R(S) with representation A = ni=1 Ai
nunion of pairwise disjoint elements of S. Show that
as a finite
μ̄(A) = i=1 μ(Ai )
2. Show that if μ̄ : R(S) → [0, +∞] is another measure with
μ̄|S = μ, i.e. another extension of μ on R(S), then μ̄ = μ̄.

www.probability.net
Tutorial 2: Caratheodory’s Extension 7

Exercise 11. Let S be a semi-ring on Ω and μ : S → [0, +∞] be a

measure. Let A be an element of R(S) with two representations:
n
 p

A= Ai = Bj
i=1 j=1

as a finite union of pairwise disjoint elements of S.

p
1. For i = 1, . . . , n, show that μ(Ai ) = j=1 μ(Ai ∩ Bj )
n p
2. Show that i=1 μ(Ai ) = j=1 μ(Bj )
3. Explain why we can define a map μ̄ : R(S) → [0, +∞] as:
n


μ̄(A) = μ(Ai )
i=1

4. Show that μ̄(∅) = 0.

www.probability.net
Tutorial 2: Caratheodory’s Extension 8

Exercise 12. Everything being as before, suppose that (An )n≥1 is

a sequence of pairwise disjoint elements of R(S), each An having the
representation:
pn
An = Akn , n ≥ 1
k=1
as a finite union of disjoint elements of S. Suppose moreover that
p
A = +∞
n=1 An is an element of R(S) with representation A = j=1 Bj ,
as a finite union of pairwise disjoint elements of S.
pn
1. Show that for j = 1, . . . , p, Bj = ∪+∞ k
n=1 ∪k=1 (An ∩ Bj ) and
+∞
explain why Bj is of the form Bj = m=1 Cm for some sequence
(Cm )m≥1 of pairwise disjoint elements of S.
+∞ pn
2. Show that μ(Bj ) = n=1 k=1 μ(Akn ∩ Bj )
3. Show that for n ≥ 1 and k = 1, . . . , pn , Akn = pj=1 (Akn ∩ Bj )

www.probability.net
Tutorial 2: Caratheodory’s Extension 9
p
4. Show that μ(Akn ) = j=1 μ(Akn ∩ Bj )
5. Recall the definition of μ̄ of exercise (11) and show that it is a
measure on R(S).

Exercise 13.Prove the following theorem:

Theorem 2 Let S be a semi-ring on Ω. Let μ : S → [0, +∞] be a

measure on S. There exists a unique measure μ̄ : R(S) → [0, +∞]
such that μ̄|S = μ.

www.probability.net
Tutorial 2: Caratheodory’s Extension 10

Definition 10 We define an outer-measure on Ω as being any

map μ∗ : P(Ω) → [0, +∞] with the following properties:
(i) μ∗ (∅) = 0
(ii) A ⊆ B ⇒ μ∗ (A) ≤ μ∗ (B)
 +∞  +∞
 
(iii) μ∗ An ≤ μ∗ (An )
n=1 n=1

Exercise 14. Show that μ∗ (A ∪ B) ≤ μ∗ (A) + μ∗ (B), where μ∗ is

an outer-measure on Ω and A, B ⊆ Ω.
Definition 11 Let μ∗ be an outer-measure on Ω. We define:

Σ(μ∗ ) = {A ⊆ Ω : μ∗ (T ) = μ∗ (T ∩ A) + μ∗ (T ∩ Ac ) , ∀T ⊆ Ω}
We call Σ(μ∗ ) the σ-algebra associated with the outer-measure μ∗ .
Note that the fact that Σ(μ∗ ) is indeed a σ-algebra on Ω, remains to
be proved. This will be your task in the following exercises.

www.probability.net
Tutorial 2: Caratheodory’s Extension 11

Exercise 15. Let μ∗ be an outer-measure on Ω. Let Σ = Σ(μ∗ ) be

the σ-algebra associated with μ∗ . Let A, B ∈ Σ and T ⊆ Ω
1. Show that Ω ∈ Σ and Ac ∈ Σ.
2. Show that μ∗ (T ∩ A) = μ∗ (T ∩ A ∩ B) + μ∗ (T ∩ A ∩ B c )
3. Show that T ∩ Ac = T ∩ (A ∩ B)c ∩ Ac
4. Show that T ∩ A ∩ B c = T ∩ (A ∩ B)c ∩ A
5. Show that μ∗ (T ∩ Ac ) + μ∗ (T ∩ A ∩ B c ) = μ∗ (T ∩ (A ∩ B)c )
6. Adding μ∗ (T ∩(A∩B)) on both sides 5., conclude that A∩B ∈ Σ.
7. Show that A ∪ B and A \ B belong to Σ.

Exercise 16. Everything being as before, let An ∈ Σ, n ≥ 1. Define

B1 = A1 and Bn+1 = An+1 \ (A1 ∪ . . . ∪ An ). Show that the Bn ’s are
pairwise disjoint elements of Σ and that ∪+∞ +∞
n=1 An = n=1 Bn .

www.probability.net
Tutorial 2: Caratheodory’s Extension 12

Exercise 17. Everything being as before, show that if B, C ∈ Σ and

B ∩ C = ∅, then μ∗ (T ∩ (B  C)) = μ∗ (T ∩ B) + μ∗ (T ∩ C) for any
T ⊆ Ω.
Exercise 18.Everything being as before, let (Bn )n≥1 be a sequence

of pairwise disjoint elements of Σ, and let B = +∞
n=1 Bn . Let N ≥ 1.

1. Explain why N
n=1 Bn ∈ Σ
N
2. Show that μ∗ (T ∩ (N
n=1 Bn )) = n=1 μ∗ (T ∩ Bn )
3. Show that μ∗ (T ∩ B c ) ≤ μ∗ (T ∩ (N c
n=1 Bn ) )

4. Show that μ∗ (T ∩ B c ) + +∞ ∗ ∗
n=1 μ (T ∩ Bn ) ≤ μ (T ), and:
+∞
5. μ∗ (T ) ≤ μ∗ (T ∩B c )+μ∗ (T ∩B) ≤ μ∗ (T ∩B c )+ n=1 μ∗ (T ∩Bn )
+∞
6. Show that B ∈ Σ and μ∗ (B) = n=1 μ∗ (Bn ).
7. Show that Σ is a σ-algebra on Ω, and μ∗|Σ is a measure on Σ.

www.probability.net
Tutorial 2: Caratheodory’s Extension 13

Theorem 3 Let μ∗ : P(Ω) → [0, +∞] be an outer-measure on Ω.

Then Σ(μ∗ ), the so-called σ-algebra associated with μ∗ , is indeed a
σ-algebra on Ω and μ∗|Σ(µ∗ ) , is a measure on Σ(μ∗ ).

Exercise 19. Let R be a ring on Ω and μ : R → [0, +∞] be a

measure on R. For all T ⊆ Ω, define:
+∞

 
∗
μ (T ) = inf μ(An ) , (An ) is an R-cover of T
n=1

where an R-cover of T is defined as any sequence (An )n≥1 of elements


of R such that T ⊆ ∪+∞
n=1 An . By convention inf ∅ = +∞.

1. Show that μ∗ (∅) = 0.

2. Show that if A ⊆ B then μ∗ (A) ≤ μ∗ (B).

www.probability.net
Tutorial 2: Caratheodory’s Extension 14

3. Let (An )n≥1 be a sequence of subsets of Ω, with μ∗ (An ) < +∞

for all n ≥ 1. Given  > 0, show that for all n ≥ 1, there exists
an R-cover (Apn )p≥1 of An such that:
+∞

μ(Apn ) < μ∗ (An ) + /2n
p=1

Why is it important to assume μ∗ (An ) < +∞.

4. Show that there exists an R-cover (Rk ) of ∪+∞
n=1 An such that:
+∞
 +∞ 
 +∞
μ(Rk ) = μ(Apn )
k=1 n=1 p=1

+∞
5. Show that μ∗ (∪+∞
n=1 An ) ≤  + n=1 μ∗ (An )
6. Show that μ∗ is an outer-measure on Ω.

www.probability.net
Tutorial 2: Caratheodory’s Extension 15

Exercise 20. Everything being as before, Let A ∈ R. Let (An )n≥1

be an R-cover of A and put B1 = A1 ∩ A, and:

Bn+1 = (An+1 ∩ A) \ ((A1 ∩ A) ∪ . . . ∪ (An ∩ A))
1. Show that μ∗ (A) ≤ μ(A).
2. Show that (Bn )n≥1 is a sequence of pairwise disjoint elements
of R such that A = +∞
n=1 Bn .

3. Show that μ(A) ≤ μ∗ (A) and conclude that μ∗|R = μ.

Exercise 21. Everything being as before, Let A ∈ R and T ⊆ Ω.

1. Show that μ∗ (T ) ≤ μ∗ (T ∩ A) + μ∗ (T ∩ Ac ).
2. Let (Tn ) be an R-cover of T . Show that (Tn ∩ A) and (Tn ∩ Ac )
are R-covers of T ∩ A and T ∩ Ac respectively.
3. Show that μ∗ (T ∩ A) + μ∗ (T ∩ Ac ) ≤ μ∗ (T ).

www.probability.net
Tutorial 2: Caratheodory’s Extension 16

4. Show that R ⊆ Σ(μ∗ ).

5. Conclude that σ(R) ⊆ Σ(μ∗ ).

Exercise 22.Prove the following theorem:

Theorem 4 (Caratheodory’s extension) Let R be a ring on Ω

and μ : R → [0, +∞] be a measure on R. There exists a measure
μ : σ(R) → [0, +∞] such that μ|R = μ.

Exercise 23. Let S be a semi-ring on Ω. Show that σ(R(S)) = σ(S).

Exercise 24.Prove the following theorem:

Theorem 5 Let S be a semi-ring on Ω and μ : S → [0, +∞] be a

measure on S. There exists a measure μ : σ(S) → [0, +∞] such that
μ|S = μ.

www.probability.net